Skip to navigation

Taxonomy and Boundaries

View as Markdown

These definitions identify the required, optional, industry-specific, NVIDIA-owned, and partner-supplied parts of an AI Factory decision. Review these boundaries before you select products. Record the result in the deployment decision record.

Required Terms

Use the following terms consistently throughout the design:

TermMeaningRequired Content
NVIDIA Enterprise AI Factory Reference DesignImplementation-flexible software architecture spanning workloads, AI platform, platform services, cloud-native platform, security, and operations on NVIDIA Enterprise Hardware Reference Architectures.Purpose, scope, layers, interfaces, decisions, non-goals, security, and operations.
NVIDIA Enterprise Hardware Reference ArchitectureAuthoritative infrastructure recommendations for systems, networking, storage, scale, and facilities.Published hardware architecture, design points, constraints, and review criteria.
Capability Reference DesignOptional architectural pattern that extends or specializes named core layers.Relationship to core, affected layers, interfaces, dependencies, non-goals, security, deployment patterns, and operations.
Industry ProfileSector-specific changes to the core.Inherited elements, extensions, additions, exclusions, sources, and accountable reviewers.
Reference implementationConcrete implementation of a Capability Reference Design on a named partner platform.Publisher, platform, applicability, assumptions, authoritative source, and content owner.
BlueprintReusable workflow, code, models, and deployment guidance for an AI use case.Release, dependencies, deployment path, and evaluation method.

Boundary Rules

The software design states the infrastructure capabilities it consumes. Enterprise Hardware Reference Architectures remain authoritative for the bill of materials, topology, system qualification, scale, and facility details.

A Capability Reference Design identifies affected layers and stable interfaces. It does not replace the core or silently redefine unrelated layers.

A profile inherits the core and records only sector-specific changes. Common platform, hardware, observability, and operations guidance stays in the core.

A partner implementation demonstrates one way to realize a capability. Inclusion does not make that implementation the universal architecture or imply certification, support equivalence, or regulatory approval.

Composition

Composition means starting with the required software design and enterprise hardware foundation, then adding only the Capability Reference Designs and Industry Profile needed for the deployment.

Industry Profile

Optional · one primary profile. Applies sector-specific additions, extensions, and exclusions to the core.

Capability Reference Designs

Optional · zero or more patterns. Adds bounded capabilities such as Secure Agent Workspace or Inference Hub.

NVIDIA Enterprise AI Factory Reference Design

Required · one software core. Defines stable layers, interfaces, security boundaries, and operational responsibilities.

Enterprise Hardware Reference Architecture

Required · one authoritative foundation. Supplies systems, networking, storage, topology, scale, and facility design points.

Composition Result

The deploying organization selects a concrete implementation that satisfies the chosen profile, capability patterns, software core, and hardware foundation.

Apply only one Industry Profile as the primary sector context. You can combine multiple Capability Reference Designs when their dependencies and security boundaries remain compatible.

Claims

Describe architecture intent, requirements, assumptions, and boundaries precisely. Do not convert the existence of a reference design, certification program, partner implementation, product support matrix, or compliance control into a broader deployment claim.